CN113361610A - Intelligent identification method and system for wine production place - Google Patents

Intelligent identification method and system for wine production place Download PDF

Info

Publication number
CN113361610A
CN113361610A CN202110646730.9A CN202110646730A CN113361610A CN 113361610 A CN113361610 A CN 113361610A CN 202110646730 A CN202110646730 A CN 202110646730A CN 113361610 A CN113361610 A CN 113361610A
Authority
CN
China
Prior art keywords
wine
prediction model
spectrogram
training
algorithm
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110646730.9A
Other languages
Chinese (zh)
Other versions
CN113361610B (en
Inventor
韦海成
塔娜
陈涛
吕新宇
蒋艳煜
肖明霞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North Minzu University
Original Assignee
North Minzu University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by North Minzu University filed Critical North Minzu University
Priority to CN202110646730.9A priority Critical patent/CN113361610B/en
Publication of CN113361610A publication Critical patent/CN113361610A/en
Application granted granted Critical
Publication of CN113361610B publication Critical patent/CN113361610B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/3577—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing liquids, e.g. polluted water
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/359—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using near infrared light
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2201/00—Features of devices classified in G01N21/00
    • G01N2201/12—Circuits of general importance; Signal processing
    • G01N2201/129—Using chemometrical methods
    • G01N2201/1296—Using chemometrical methods using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Abstract

The invention relates to an intelligent identification method and system for a wine production place, wherein the method comprises the following steps: collecting spectral data of the wine to be detected and drawing a spectrogram; preprocessing the spectrogram, wherein the preprocessing comprises dimensionality reduction and denoising; extracting characteristic variables from the preprocessed spectrogram by using a pre-trained prediction model, classifying the extracted characteristic variables, and determining the production area of the wine based on the classification of each characteristic variable; the prediction model is an RBM (radial basis function) series multilayer BP (back propagation) neural network model for support vector machine analysis, and is obtained by training a plurality of RBMs in a multilayer series connection mode as a forward algorithm, a back update algorithm adopts a BP algorithm and is fused with a support vector machine algorithm. Different from the traditional near infrared spectrum detection, the method combines the chemical detection and the machine learning technology, and can eliminate the influence of noise existing in the traditional spectrum detection, thereby improving the accuracy of the wine production place identification.

Description

Intelligent identification method and system for wine production place
Technical Field
The invention relates to the technical field of food detection, in particular to an intelligent identification method and system for a wine production place.
Background
With the improvement of the living standard of people, the consumption of the wine is continuously increased, and the wine with the geographical sign is widely accepted by the market because the wine is a food with strong regionality. Under the background, in order to obtain improper benefits, a great deal of counterfeit and shoddy wine is produced, and wine protected by a non-geographic sign is often sold as wine protected by a geographic sign. In order to improve the brand effect and the economic benefit of the geographical mark protection wine and maintain the legal rights and interests of consumers, the intelligent identification technology of the origin of the wine is of great importance.
At present, various detection technologies are applied to the identification of the production place of the wine at home and abroad, and the wine is mainly analyzed from the aspects of phenolic substances, amino acids, volatile components, trace elements, isotopes, sensory characteristics and the like, and the original production place of the wine is traced by combining a chemometric method. Spectroscopic analysis is a rapid and sensitive detection technique that utilizes the emission, absorption or scattering spectra of chemical substances (atoms, groups, molecules and high molecular compounds) to determine their properties, structures or contents. However, the visible/near infrared spectrum technology is greatly interfered by external environment, so that infrared spectrum information contains a lot of noises irrelevant to sample information, and when the visible/near infrared spectrum is used for modeling, the modeling efficiency and the prediction performance of a model are influenced.
Disclosure of Invention
The invention aims to overcome the defects that modeling efficiency and predictive performance are easily influenced by irrelevant noise during visible/near infrared spectrum detection in the prior art, and provides an intelligent identification method and system for a wine production place.
In order to achieve the above object, the embodiments of the present invention provide the following technical solutions:
on one hand, the embodiment of the invention provides an intelligent identification method for a wine production place, which comprises the following steps:
collecting spectral data of the wine to be detected and drawing a spectrogram;
preprocessing the spectrogram, wherein the preprocessing comprises dimensionality reduction and denoising;
extracting characteristic variables from the preprocessed spectrogram by using a pre-trained prediction model, classifying the extracted characteristic variables, and determining the production area of the wine based on the classification of each characteristic variable;
the prediction model is an RBM (radial basis function) series multilayer BP (back propagation) neural network model for support vector machine analysis, and is obtained by training a plurality of RBMs in a multilayer series connection mode as a forward algorithm, a back update algorithm adopts a BP algorithm and is fused with a support vector machine algorithm.
In a more refined approach, the prediction model is trained by:
s201, collecting a plurality of wine samples of different producing areas, and randomly dividing the wine samples into a training set and a testing set;
s202, aiming at a training set and a testing set, collecting spectral data of a wine sample, drawing a spectrogram, and preprocessing the spectrogram;
s203, extracting characteristic variables and predicting and classifying the spectrogram after the preprocessing in the training set by using the initially established prediction model;
s204, detecting phenolic components and anthocyanin components in the wine by adopting a quadrupole time-of-flight mass spectrometer;
s205, correcting the classification of the characteristic variables obtained in the step S203 through the detection result obtained in the step S204, optimizing the prediction model to obtain an optimized prediction model, then returning to the step S203, and in the step S203, re-extracting the characteristics and the prediction classification of the original spectrogram by using the optimized prediction model;
and (5) circularly executing the steps S203-S205 until the training is finished.
On the other hand, the embodiment of the invention also provides an intelligent identification system for a wine production place, which comprises the following components:
the spectrogram drawing module is used for collecting spectral data of the wine to be detected and drawing a spectrogram;
the preprocessing module is used for preprocessing the spectrogram, and the preprocessing comprises denoising and dimensionality reduction;
the producing area identification module is used for extracting characteristic variables from the preprocessed spectrogram by using a pre-trained prediction model, classifying the extracted characteristic variables and determining the producing area of the wine based on the classification of the characteristic variables;
the prediction model is an RBM (radial basis function) series multilayer BP (back propagation) neural network model for support vector machine analysis, and is obtained by training a plurality of RBMs in a multilayer series connection mode as a forward algorithm, a back update algorithm adopts a BP algorithm and is fused with a support vector machine algorithm.
In a further aspect, the present invention provides a computer-readable storage medium of computer-readable instructions, which when executed, cause a processor to perform the operations of the intelligent wine production area identification method according to any one of the embodiments of the present invention.
In another aspect, an embodiment of the present invention provides an electronic device, including: a memory storing program instructions; and the processor is connected with the memory and executes the program instructions in the memory to realize the intelligent identification method of the wine production place in any embodiment of the invention.
Compared with the prior art, the method has the following technical advantages:
in the invention, the visible/near infrared spectrum data of the wine are classified and predicted by adopting an algorithm of an RBM (radial basis function) series multilayer BP (Back propagation) neural network architecture which is analyzed by a fusion support vector machine, and the unique components of the wine in different production areas are searched, so that the production area identification is realized. The specific method comprises the following steps: firstly, feature extraction and classification are carried out on visible/near infrared spectrum data of the wine by adopting an RBM serial multilayer BP neural network architecture. Secondly, the detection model established by the visible/near infrared spectrum can be corrected by the Q-TOF and other chemical metering means, and the chemical detection and the spectrum detection result are fused and corrected by the deep learning technology, so that the detection efficiency and the detection precision are improved.
Different from the traditional near infrared spectrum variable screening and correcting model, the method is more scientific and accurate. In the aspect of detection, the visible/near infrared spectrum technology is simpler and more convenient than the chromatographic technology, and can carry out real-time detection.
Other advantages of the invention will be apparent from the detailed description which follows.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
Fig. 1 is a flowchart illustrating an intelligent identification method of a wine production area in an embodiment.
FIG. 2 is a flowchart of a method for training the prediction model according to an embodiment.
Fig. 3 is a basic structural diagram of an RBM.
FIG. 4 is a schematic structural diagram of the prediction model in the embodiment.
FIGS. 5a, 5b, 5c and 5d are graphs showing the results of the detection of the origin of wine in examples.
FIG. 6 is a functional block diagram of an intelligent identification system of wine origin in the embodiment.
FIG. 7 is a functional block diagram of a model training module.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, the present embodiment provides an intelligent identification method for a wine production area, including the following steps:
and S10, collecting the spectral data of the wine to be detected and drawing a spectrogram.
The wine to be detected is the wine whose origin is to be identified, and the visible/near infrared spectrum optical fiber acquisition equipment is used for acquiring the spectrum data of the wine stock solution, and then the spectrum diagram of the wine stock solution is drawn. Different regions of the spectrogram contain various types of information, and each light absorption peak has a corresponding functional group corresponding to different components. The acquisition of spectral data and the plotting of spectral plots are prior art and will not be described in detail in this embodiment.
Experiments show that the wines in different production areas have no obvious difference in peak positions, but have obvious difference in light absorption intensity, so that the production areas can be identified by spectrograms theoretically. However, the real application process finds that the visible/near infrared spectrum has the problem of aliasing of component or functional group absorption intensity, and then the identification effect is influenced, so that the production place identification is not directly carried out based on the spectrogram in the embodiment, but characteristic variables are extracted from the spectrogram, and then the production place identification is carried out based on the characteristic variables and matched with a machine learning method.
S20, preprocessing the spectrogram, wherein the preprocessing comprises denoising and denoising.
The spectrogram is preprocessed before extracting the characteristic variables so as to improve the quality of the spectrogram and then extract the characteristic variables better. The preprocessing in this embodiment includes data dimension reduction and denoising.
For example, by performing smoothing filtering by using a savitzky golay (S-G for short), the smoothness of the spectrum can be improved, and the noise interference can be reduced. The effect of the S-G smoothing filter varies with the selected window width. Assume a window size of 5, i.e., 5 points at a time, including itself and 2 points before and after. Taking 5 points of a spectrum in a segment of the spectrum at equal wavelength intervals as an X set, the polynomial smoothing is based on the fact that X is taken as a wavelength pointk−2,Xk−1,Xk,Xk+1,Xk+2Replacing X by a polynomial fit value of the data ofkAnd k represents the kth point on the trace point, and then the points are sequentially moved until the spectrum is traversed.
For example, a Principal Component Analysis (PCA) method can be used for data dimensionality reduction. The PCA algorithm finds a group of vectors in a data space, expresses the variance of data as much as possible by using the vectors, reduces the data from a high dimension to a low dimension, obtains a lowest dimension identification space which is close to the original data space by using K-L transformation, and enables the projection rear directions of the lowest dimension identification space to be orthogonal with each other. And the sizes of the main components are sequentially extracted according to the size of the variance. The method is an unsupervised dimension reduction algorithm, the data category attribute is not required to be considered, comprehensive transformation is obtained through certain linear projection change, and the comprehensive variable reflects the rough rule reflected in the original spectral data to the maximum extent. The feature selection of data dimension reduction needs to select variables by minimizing an objective function to construct a sparse dictionary. And then, the real-time deep learning model is introduced into the spectral feature field, so that the feature identification of the nonlinear function among the high-dimensional features is realized.
And S30, inputting the preprocessed spectrogram into a pre-trained prediction model, and outputting to obtain a recognition result. That is, feature variables are extracted from the spectrogram by using a pre-trained prediction model, the extracted feature variables are classified, and the production place of the wine is determined based on the classification of each feature variable.
The prediction model is a pre-trained model, and is an RBM (radial basis function) series multilayer BP (back propagation) neural network model supporting vector machine analysis, namely a model obtained by combining three technologies of support vector machine analysis, multilayer RBM series and BP neural network training together.
Referring to fig. 2, the training process of the prediction model includes the following steps:
s201, collecting a plurality of wine samples of different producing areas, wherein the wine samples of the same producing area can be in multiple parts, and randomly dividing the wine samples into a training set and a testing set.
When collecting samples, it is better to collect samples of each production place as many as possible, so that the trained prediction model can well identify the wine of each production place.
S202, aiming at the training set and the testing set, collecting the spectrum data of the wine sample, drawing a spectrogram, and preprocessing the spectrogram. The preprocessing includes denoising and denoising, and the specific implementation can refer to the relevant description in step S20.
And S203, extracting characteristic variables and predicting and classifying the spectrogram after the preprocessing in the training set by using the initially established prediction model.
The basic structure of an RBM (Restricted Boltzmann Machine, for short) is shown in fig. 3, and a neuron is a two-layer neural network, a first layer is a visible layer, and a second layer is a hidden layer. Since there is no inter-level communication, it is not much different from a full connection in the forwarding process, but there is actually a reverse process, which is an undirected graph. The RBM reconstructs the data by itself in an unsupervised manner, in the reconstruction phase the activation state of the hidden layer is multiplied by the same weight for each connected edge, the sum of these products is added at each visible node to the bias term of the visible layer, each reconstruction process being an approximation to the original input. The bottom visual layer of the model is an observable variable, and the top connection is undirected, so that the state of an unknown variable is deduced under the observable variable, and the hidden state is adjusted to reconstruct observable data.
As shown in fig. 4, in this embodiment, the structure of the RBMs is improved, a plurality of RBMs are connected in series in multiple layers to serve as a forward algorithm, a later update algorithm is a BP algorithm, and a support vector machine algorithm is fused, so that an RBM series multilayer BP neuron network model for support vector machine analysis is provided. Pre-training by using a layer-by-layer unsupervised method, training only one RBM at each time, namely training the structures of two adjacent layers, and classifying by using a support vector machine algorithm at the top layer after training of all the layers is finished. In the aspect of classifier integration, an improved Boosting algorithm cluster serial algorithm is selected, and is an Adaboost algorithm, and the algorithm principle is that a weak classifier with the minimum weight coefficient is screened from trained weak classifiers by adjusting sample weights and weak classifier weights to form a final strong classifier. The 0/1 loss function is replaced by an exponential loss function with better mathematical property, more samples which are wrongly classified are concerned, weak classifiers with good selective performance are emphasized, the samples which are wrongly classified are endowed with larger weight values in each circulation, the weight values of the samples which are correctly classified are smaller, one weak classifier is generated in each circulation, the weight value of each classifier is adjusted, and finally, the classifiers are integrated by adopting a weighted voting method to obtain the classifier with better classification effect. The classification accuracy can be better improved by the model.
More specifically, this step includes the following processing steps:
(1) pre-training a single-layer RBM, and then stacking multiple layers to form a network.
Inputting the preprocessed spectral data, and separately and unsupervised training each layer of RBM network respectively to ensure that the feature vectors are mapped to different feature spaces and feature information is kept as much as possible; weights are obtained through unsupervised greedy layer-by-layer method pre-training, class labels are not needed, fitting input is conducted continuously, and the weights are sequentially added layer by layer (namely, each layer of RBM is pre-trained independently and then overlapped layer by layer sequentially). In the process, data passes through the visible layer to generate a vector V, and then the vector V is transmitted to the hidden layer by the weight w to obtain h. The stability of the network is measured through an energy function, and the optimization function is obtained through solving an index according to the energy function, normalizing and then obtaining the maximum likelihood. For example, in single-layer RBM training, the visible layer is used to receive the input signal, and the hidden layer is used to extract features.
(2) A BP network is arranged at the last layer of the network (namely, the network formed by overlapping a plurality of layers of RBM networks), the output feature vector of the RBM is received as the input feature vector of the RBM, and an entity relationship classifier is supervised trained (the entity relationship classifier is a classifier for classifying the extracted feature vector). Each layer of RBM network can only ensure that the weight in the layer of the RBM network can be optimal for the mapping of the characteristic vector of the layer, but not optimal for the mapping of the characteristic vector of the whole multilayer RBM network, so the back propagation network also propagates error information to each layer of RBM from top to bottom, and the whole multilayer RBM network is finely adjusted. The process of the RBM network training model can be regarded as the initialization of a deep BP network weight parameter, so that the multilayer RBM network overcomes the defects that the BP network is easy to fall into local optimum and the training time is long due to the random initialization of the weight parameter, the modeling efficiency can be accelerated, and the prediction precision can be improved.
(3) The model algorithm is fused with the SVM and Boosting algorithm, so that the classification precision and the generalization capability are improved. The SVM + Boosting algorithm combines Boosting and an SVM algorithm, a small training sample is used for training the SVM to obtain a classifier, and the Boosting method is further used for improving the generalization capability of the SVM. The BP network is based on the traditional statistics, the content of the traditional statistics research is the progressive theory when the sample is infinite, namely the sample data is always limited in the practical problem when the sample data tends to be infinite. The fusion SVM + Boosting algorithm can overcome the problem that a BP network is difficult to avoid, and has strong approximation capability and generalization capability.
And S204, detecting phenolic components and anthocyanin components in the wine by adopting Q-TOF (quadrupole time of flight mass spectrometer).
Q-TOF chemical detection is prior art, samples are pre-prepared (0.22 micron filtration wine stock) and are provided with a mobile phase (methanol, 5mmol ammonium formate), and chromatographic and mass spectrometric conditions are respectively set for testing. The primary mass spectrum MS result is used for searching for the difference components, and the secondary mass spectrum MS/MS is used for verifying the difference components. Since Q-TOF is prior art, it is not described in detail here.
In the experiment, the test results are shown in fig. 5a, 5b, 5c and 5d, taking three producing areas of wine, namely kanglan, red temple and yinchuan as an example. FIG. 5a is a graph of the near infrared spectra of wines from the three production areas of Nakan, Hongyinburg and Yinchuan, which show slight differences in reflectance although the wave numbers corresponding to the peak positions are similar. This difference appears more clearly in the fingerprint region. FIGS. 5b, 5c and 5d are Q-TOF metabolome components of three-region wine, wherein 15 kinds of wines (3-pyridylpropionic acid, quinine, lidocaine, procainamide, progesterone, acephate tetraacetic acid, cortisone, gemfibrozil, 4-aminobenzoate, meprobamate, albuterol, chenodeoxycholic acid, labetalol, acetazolamide and norethindrone) are provided in the area of production of Gantang, and 19 kinds of wines (3-pyridylpropionic acid, lidocaine, procainamide, progesterone, acephate tetraacetic acid, cortisone, gemfibrozil, indole, hexyl butyl, 4-aminobenzoate, meprobamate, albuterol, chenodeoxycholic acid, labetalol, acetazolamide, norethindrone, reserpine, terbutaline and phenylbutanone) are provided in the area of production of Gantang; there are 5 different components (quinine, indole, reserpine, terbutaline and phenylbutanone) in wine in Yinchuan and Red Tembo producing areas.
S205, the classification of the characteristic variables obtained in the step S203 is corrected according to the detection result obtained in the step S204, the prediction model is optimized to obtain an optimized prediction model, then the step S203 is returned, and in the step S203, the characteristics and the prediction classification of the original spectrogram are re-extracted by using the optimized prediction model. Steps S203-S205 are executed in a loop until the training is finished, i.e. until the prediction model converges to the local optimum point.
It should be noted here that since the detection result of step S204 is the same during each loop, step S204 may be preferably executed only once, and the detection result may be called directly during each loop, so as to reduce the amount of computation.
During correction, a near infrared spectrum wave number range corresponding to the difference components detected and displayed by the Q-TOF is searched, whether the characteristic wave number range is consistent with the characteristic range extracted by the prediction model or not is compared, if not, parameters of the prediction model are adjusted or a hidden layer is added, and the purpose of model correction is achieved. And if the model structure is consistent with the current model structure, keeping the current model structure.
And S206, testing the prediction model obtained after the training is finished by using the test set, and verifying the accuracy of the prediction model. This step may be an optional step in training.
Referring to fig. 6 and 7, the present embodiment provides an intelligent identification system for a wine production area, including:
the model training module is used for training to obtain a prediction model, the prediction model is an RBM (radial basis function) series multilayer BP (back propagation) neural network model supporting vector machine analysis, the prediction model is obtained by performing multilayer series connection on a plurality of RBMs as a forward algorithm, adopting a BP algorithm as a posterior updating algorithm and fusing the algorithm of the support vector machine. Since the model training module is not needed in the application, but only needed before the application, it is represented in fig. 6 by a dashed box.
And the spectrogram drawing module is used for collecting spectral data of the wine to be detected and drawing a spectrogram.
And the preprocessing module is used for preprocessing the spectrogram, and the preprocessing comprises denoising and denoising.
And the producing area identification module is used for extracting characteristic variables from the preprocessed spectrogram by using a pre-trained prediction model, classifying the extracted characteristic variables and determining the producing area of the wine based on the classification of the characteristic variables.
As shown in fig. 7, the model training module includes:
the sample collection submodule is used for collecting a plurality of wine samples of different producing areas and randomly dividing the wine samples into a training set and a testing set;
the pretreatment submodule is used for acquiring spectral data of the wine sample aiming at the training set and the testing set, drawing a spectrogram and pretreating the spectrogram;
the prediction submodule is used for extracting characteristic variables and predicting and classifying the spectrogram after the initial set-up prediction model is used for preprocessing in the training set;
the detection submodule is used for detecting phenolic components and anthocyanin components in the wine by adopting a quadrupole time-of-flight mass spectrometer;
the correction submodule corrects the classification of the characteristic variables output by the predictor submodule according to the obtained detection result and optimizes the prediction model;
and the end judgment submodule is used for judging whether the training is ended, if not, the prediction submodule is driven to re-extract the characteristics and predict and classify the original spectrogram by using the optimized prediction model, and the correction submodule is driven to optimize the prediction model until the training is judged to be ended.
The intelligent identification system of wine origin is proposed based on the same inventive concept of wine origin intelligent identification, and for the unclear part of the system, please refer to the related description in the foregoing method embodiment.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. An intelligent identification method for a wine production place is characterized by comprising the following steps:
collecting spectral data of the wine to be detected and drawing a spectrogram;
preprocessing the spectrogram, wherein the preprocessing comprises dimensionality reduction and denoising;
extracting characteristic variables from the preprocessed spectrogram by using a pre-trained prediction model, classifying the extracted characteristic variables, and determining the production area of the wine based on the classification of each characteristic variable;
the prediction model is an RBM (radial basis function) series multilayer BP (back propagation) neural network model for support vector machine analysis, and is obtained by training a plurality of RBMs in a multilayer series connection mode as a forward algorithm, a back update algorithm adopts a BP algorithm and is fused with a support vector machine algorithm.
2. An intelligent identification method of wine origin according to claim 1, characterized in that the predictive model is trained by the following steps:
s201, collecting a plurality of wine samples of different producing areas, and randomly dividing the wine samples into a training set and a testing set;
s202, aiming at a training set and a testing set, collecting spectral data of a wine sample, drawing a spectrogram, and preprocessing the spectrogram;
s203, extracting characteristic variables and predicting and classifying the spectrogram after the preprocessing in the training set by using the initially established prediction model;
s204, detecting phenolic components and anthocyanin components in the wine by adopting a quadrupole time-of-flight mass spectrometer;
s205, correcting the classification of the characteristic variables obtained in the step S203 through the detection result obtained in the step S204, optimizing the prediction model to obtain an optimized prediction model, then returning to the step S203, and in the step S203, re-extracting the characteristics and the prediction classification of the original spectrogram by using the optimized prediction model;
and (5) circularly executing the steps S203-S205 until the training is finished.
3. The intelligent identification method for wine production place according to claim 2, wherein the step S203 comprises the steps of:
(1) inputting the preprocessed spectral data for pre-training, separately and unsupervised training each layer of RBM network, and then overlapping the multiple layers of RBM networks;
(2) setting a BP network at the last layer of the multilayer RBM network, receiving an output feature vector of the RBM as an input feature vector, and training a classifier in a supervision manner;
(3) and training a classifier by combining Boosting and SVM algorithms to obtain the prediction model.
4. The intelligent wine production area identification method according to claim 2, wherein in the step S205, during the correction, a near infrared spectrum wave number range corresponding to the difference component detected and displayed by the quadrupole time-of-flight mass spectrometer is searched, whether the near infrared spectrum wave number range is consistent with the extraction characteristic range of the prediction model or not is compared, and if the near infrared spectrum wave number range is not consistent with the extraction characteristic range of the prediction model, the prediction model is adjusted in parameters or a hidden layer is added, so as to realize the correction of the prediction model.
5. The intelligent identification method of wine origin according to claim 2, further comprising, after step S205, the steps of: and S206, testing the prediction model obtained after the training is finished by using the test set so as to verify the accuracy of the prediction model.
6. An intelligent identification system for a wine production place, comprising:
the spectrogram drawing module is used for collecting spectral data of the wine to be detected and drawing a spectrogram;
the preprocessing module is used for preprocessing the spectrogram, and the preprocessing comprises denoising and dimensionality reduction;
the producing area identification module is used for extracting characteristic variables from the preprocessed spectrogram by using a pre-trained prediction model, classifying the extracted characteristic variables and determining the producing area of the wine based on the classification of the characteristic variables;
the prediction model is an RBM (radial basis function) series multilayer BP (back propagation) neural network model for support vector machine analysis, and is obtained by training a plurality of RBMs in a multilayer series connection mode as a forward algorithm, a back update algorithm adopts a BP algorithm and is fused with a support vector machine algorithm.
7. A wine production place intelligent recognition system as claimed in claim 6, further comprising a model training module for training to obtain the prediction model; the model training module comprises:
the sample collection submodule is used for collecting a plurality of wine samples of different producing areas and randomly dividing the wine samples into a training set and a testing set;
the pretreatment submodule is used for acquiring spectral data of the wine sample aiming at the training set and the testing set, drawing a spectrogram and pretreating the spectrogram;
the prediction submodule is used for extracting characteristic variables and predicting and classifying the spectrogram after the initial set-up prediction model is used for preprocessing in the training set;
the detection submodule is used for detecting phenolic components and anthocyanin components in the wine by adopting a quadrupole time-of-flight mass spectrometer;
the correction submodule corrects the classification of the characteristic variables output by the prediction submodule according to the obtained detection result and optimizes the prediction model;
and the end judgment submodule is used for judging whether the training is ended, if not, the prediction submodule is driven to re-extract the characteristics and predict and classify the original spectrogram by using the optimized prediction model, and the correction submodule is driven to optimize the prediction model until the training is judged to be ended.
8. The intelligent wine production area identification system of claim 7, wherein the correction sub-module searches for a near infrared spectrum wave number range corresponding to the difference component detected and displayed by the quadrupole time-of-flight mass spectrometer during correction, compares whether the near infrared spectrum wave number range is consistent with the extraction characteristic range of the prediction model, and adjusts parameters of the prediction model or adds a hidden layer if the near infrared spectrum wave number range is inconsistent with the extraction characteristic range of the prediction model, so as to correct the prediction model.
9. A computer readable storage medium of computer readable instructions, wherein the computer readable instructions, when executed, cause a processor to perform the operations of the method of any of claims 1-5.
10. An electronic device, comprising:
a memory storing program instructions;
a processor coupled to the memory and executing the program instructions in the memory to implement the method of any of claims 1-5.
CN202110646730.9A 2021-06-10 2021-06-10 Intelligent identification method and system for wine production place Active CN113361610B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110646730.9A CN113361610B (en) 2021-06-10 2021-06-10 Intelligent identification method and system for wine production place

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110646730.9A CN113361610B (en) 2021-06-10 2021-06-10 Intelligent identification method and system for wine production place

Publications (2)

Publication Number Publication Date
CN113361610A true CN113361610A (en) 2021-09-07
CN113361610B CN113361610B (en) 2023-01-31

Family

ID=77533638

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110646730.9A Active CN113361610B (en) 2021-06-10 2021-06-10 Intelligent identification method and system for wine production place

Country Status (1)

Country Link
CN (1) CN113361610B (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090246427A1 (en) * 2008-03-28 2009-10-01 Hincks Daniel A Product labels having removable portions having adhesive and backing thereon
CN103235057A (en) * 2013-04-27 2013-08-07 江南大学 Method for identifying white spirit origin place by using gas phase chromatography-mass spectrometry without analyzing compounds
CN105044198A (en) * 2015-07-03 2015-11-11 中国农业大学 Mineral element-based fingerprint identification method used for identifying wine countries of origin
CN106560841A (en) * 2016-10-20 2017-04-12 中国计量大学 Wuyi rock tea production place identification method based on deep learning
CN107219188A (en) * 2017-06-02 2017-09-29 中国计量大学 A kind of method based on the near-infrared spectrum analysis textile cotton content for improving DBN
CN108489927A (en) * 2018-01-24 2018-09-04 仲恺农业工程学院 Fish place of production source tracing method, electronic equipment, storage medium and device
CN109520962A (en) * 2017-09-19 2019-03-26 江南大学 A kind of grape wine near infrared spectrum detection method
CN111563558A (en) * 2020-05-13 2020-08-21 宿州学院 Rapid identification method for producing area and brand of wine

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090246427A1 (en) * 2008-03-28 2009-10-01 Hincks Daniel A Product labels having removable portions having adhesive and backing thereon
CN103235057A (en) * 2013-04-27 2013-08-07 江南大学 Method for identifying white spirit origin place by using gas phase chromatography-mass spectrometry without analyzing compounds
CN105044198A (en) * 2015-07-03 2015-11-11 中国农业大学 Mineral element-based fingerprint identification method used for identifying wine countries of origin
CN106560841A (en) * 2016-10-20 2017-04-12 中国计量大学 Wuyi rock tea production place identification method based on deep learning
CN107219188A (en) * 2017-06-02 2017-09-29 中国计量大学 A kind of method based on the near-infrared spectrum analysis textile cotton content for improving DBN
CN109520962A (en) * 2017-09-19 2019-03-26 江南大学 A kind of grape wine near infrared spectrum detection method
CN108489927A (en) * 2018-01-24 2018-09-04 仲恺农业工程学院 Fish place of production source tracing method, electronic equipment, storage medium and device
CN111563558A (en) * 2020-05-13 2020-08-21 宿州学院 Rapid identification method for producing area and brand of wine

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
VINCENT: "RBM（限制玻尔兹曼机）、DBN（深度信念网络）介绍", 《CSDN》 *
王静等: "基于深度信念网络的烟叶部位近红外光谱分类方法研究", 《红外与激光工程》 *

Also Published As

Publication number Publication date
CN113361610B (en) 2023-01-31

Similar Documents

Publication Publication Date Title
Singh et al. Comparative study ID3, cart and C4. 5 decision tree algorithm: a survey
US20220254350A1 (en) Method, apparatus and device for voiceprint recognition of original speech, and storage medium
CN109934269B (en) Open set identification method and device for electromagnetic signals
EA036070B1 (en) Optical fiber perimeter intrusion signal identification method, identification device and optical fiber perimeter intrusion alarm system
Luinge Automated interpretation of vibrational spectra
CN108766464B (en) Digital audio tampering automatic detection method based on power grid frequency fluctuation super vector
WO2022257458A1 (en) Vehicle insurance claim behavior recognition method, apparatus, and device, and storage medium
CN112613536A (en) Near infrared spectrum diesel grade identification method based on SMOTE and deep learning
CN115358481A (en) Early warning and identification method, system and device for enterprise ex-situ migration
CN113408616B (en) Spectral classification method based on PCA-UVE-ELM
CN113361610B (en) Intelligent identification method and system for wine production place
CN111666999A (en) Remote sensing image classification method
CN111398238A (en) Laser-induced fluorescence spectrum identification method for edible oil doped with castor oil
CN114863286B (en) Mixed waste plastic classification method based on multi-algorithm collaborative optimization
CN114062306B (en) Near infrared spectrum data segmentation preprocessing method
CN109408498A (en) The identification of time series feature and decomposition method based on eigenmatrix decision tree
US11525774B2 (en) Sensory evaluation method for spectral data of mainstream smoke
Ma The Research of Stock Predictive Model based on the Combination of CART and DBSCAN
CN112326574A (en) Spectrum wavelength selection method based on Bayesian classification
CN112232329A (en) Multi-core SVM training and alarming method, device and system for intrusion signal recognition
CN114694771A (en) Sample classification method, training method of classifier, device and medium
CN114826764B (en) Edge computing network attack recognition method and system based on ensemble learning
CN115908773A (en) Intelligent identification method and system for wine production place
CN117557917B (en) Water quality detection method and device
Xu et al. User Intention Prediction Method Based on Hybrid Feature Selection and Stacking Multi-model Fusion

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant